Software Based Pipeline Leak Detection Presented by James

Software - Based Pipeline Leak Detection Presented by: James Akingbola David Mannel Elijah Odusina

Overview Introductio n Methods of leak detection Hardware Software Generalize d Likelihood Ratio Simulation Procedure & Results

Introduction Pipelines used as bulk carriers of crude oil and natural gas p Used in water distribution systems p Results of Leakages include: p n n n Loss of product Environmental hazards Loss of life

Introduction Gas pipeline explosion in the Ukrainian village of Luka south of the capital Kiev May 7, 2007

Pipeline Configurations

Methods of leak detection p Hardware Method: This method is used to detect leaks with instrumentation outside the pipeline. p Software Method: This method uses different instruments to monitor internal parameters (such as pressure, flow, temperature) that infer product release. It is also known as computational pipeline monitoring (CPM).

Hardware Method p Acoustic Emissions p Fiber Optic Sensing p Vapor Monitoring p Multi-hop Sensor Wireless Network* * Method developed by OU faculty: Professor Sridhar Radhakrishnan

Hardware Method - Acoustic Emissions * http: //www. wavealert. com/Pages/sys 1. html

Hardware Method – Fiber Optic .

Hardware Method – Vapor Sensing Method

Pipeline Monitoring: Multi-hop Sensor Wireless Network Professor Sridhar Radhakrishnan Problem: Develop continuous real-time monitoring of pipelines to determine leak and other structural damages. Solution Multi-hop wireless sensor network with appropriate sensor fusion technologies. Importance: Failure of gas pipelines will result in both human, property, and environmental damage. Current Solution: Low flying aircraft, visual inspection, and use of pigs for internal monitoring. Weaknesses: Expensive, non-continuous monitoring, fail-first, fix-later solutions. School of Computer Science Research Issues • Communication in the presence of unreliable sensors • Power aware strategies • Optimal sensor configuration and data fusion

Hardware Methods p Advantages n n p Good sensitivity to leak Very accurate in leak locating Disadvantages n n System costs are usually high High complexity of installation

Software Methods p Balancing Systems n p Uses the principle of mass conservation Types of Balancing systems n n n Volume Balance Compensated Mass Balance Model Compensated Mass Balance

Software Methods- Balancing Systems p Advantage n n p It is simple to implement Cost effective Disadvantage n n Can be affected by instrument error High rate of false alarms

Software Method - Pressure Analysis n Gradient Intersection method p p This method uses SCADA* values to calculate theoretical hydraulic profile or baseline of the pipe The presence of a leak can be determined from a specific deviation or combinations of several deviations from the baseline *Supervisory Control and Data Acquisition (SCADA)

Software Method – Gradient Intersection p Advantages n p Estimation of leak location can be determined Disadvantages n n n Dependent on instrument sensitivity. Smaller leaks typically take longer time to detect. Does not give magnitude of leak

Software Method – Gradient Intersection Xleak

Software Method - RTTM p Real Time Transient Modeling (RTTM) n n Involves computer simulation of pipeline conditions using advanced fluid mechanics and hydraulic modeling. Basic equations used in RTTM are: p Continuity equation: p Momentum Equation: p Energy equation:

Software Method - RTTM p Advantages n n p It takes into account the configuration of the pipe as well as the product characteristics. Very fast in detection and location Disadvantage n n It is a complex way of leak detection The RTTM method costs much more than the other methods Requires many instruments, controller training and maintenance. Errors in instrument calibration could raise false alarms

Generalized Likelihood Ratio This is a statistical method modeled after the flow conditions in the pipeline p A mathematical model that describes effects of leaks and biases on the flow process is used. p Can detect leaks in pipeline branch, location in the branch and magnitude of the leak. p Can identify various types of gross errors p

GLR for Gross Error Identification Process Model Steady state model without leak is a measurement vector is the true value of state variables is the vector of random error Process leak Model A mass flow leak in process unit (node) j of unknown magnitude b can be modeled by; the elements of vector corresponds to the total mass flow constraint associated with node j = constraint matrix Measurement bias Model b is the bias of unknown magnitude in instrument I Procedure for single gross error When there is no gross error; =is a vector with unity in position i S. Narasimhan and R. S. H. Mah. "Generalized Likelihood Ratio Method for Gross Error Identification. " AIChe Journal 33, No. 9(1987): 1514 -1519.

GLR for Gross Error Identification If a gross error due to a bias of magnitude b is present in measurement I, then; If a gross error due to process leak in magnitude b is present in node j, then; When a gross error due to a bias or process leak is present; For a bias in measurement i For a process leak in node j let μ be the unknown expected value of r, we can formulate the hypotheses for gross error detection as Ho: is the null hypothesis that no gross errors are present and H 1: is the alternative hypothesis that either a leak or a measurement bias is present. b and fi are unknown parameters. b can be any real number and fi will be referred to as a gross error vectors from the set F

GLR for Gross Error Identification We will use the likelihood ratio test statistics to test the hypothesis by: The maximum likelihood estimate : Substituting in the test statistics equation and denoting T by Ti: The expression on the right hand side is always positive. The calculation can be simplified by the calculation by the test statistics, T as: Where: This calculation is performed for every vector fi in set F and the test statistics T is:

GLR for Gross Error Identification NOTE The gross error that corresponds to vector f* is identified as the gross error and its magnitude b^ can be estimated Performance Measures The overall power of the method to identify gross errors is given by: Overall power Number of gross errors correctly identified Number of gross errors simulated

GLR for Gross Error Identification p Results & Discussion For the Recycle process network 4 1 2 3 6 5 7

GLR for Gross Error Identification Constraint Matrix for the process Network Type of Constraint Mass Balance Streams Pipe 1 2 3 4 5 6 7 A 1 -1 0 1 0 B 0 1 -1 0 0 C 0 0 1 -1 -1 0 0 D 0 0 1 -1 -1

GLR for Gross Error Identification Bias in sensor 1 Sensors 1 2 Simulations (Ti) 1 2 2110 719 3 4 5 6 7 8 9 10 1190 431 1496 2457 1394 2386 1404 1056 3 61 82 729 631 16 227 0. 18 141 181 491 930 712 1 521 34 10 189 535 491 4 4 371 5 159 196 99 19 751 10 114 528 5 6 1144 379 317 550 270 29 86 583 283 72 1008 926 217 127 32 10 21 33 39 261 7 919 1538 0. 18 975 523 755 T 2110 729 1538 583 1496 2457 1394 2659 1404 1056 39 2659 1052 336 b^ -58 34 -50 -35 49 -63 47 66 48 -41 b -50 -35 35 -33 32 -60 45 -55 40 -50 Overall power = 0. 6

GLR for Gross Error Identification Leaks in node B and C Overall power= 0. 8

Advanced Pipe Network

Error vs. Simulated Magnitude (1% Meter Variance) 300, 0000% 250, 0000% Error 200, 0000% 150, 0000% TRUE FALSE 100, 0000% 50, 0000% 0 20 40 60 80 Simulated Magnitude 100 120 140

Error vs Simulated Magnitude (3% Meter Variance) 300, 0000% 250, 0000% Error 200, 0000% 150, 0000% TRUE FALSE 100, 0000% 50, 0000% -250 -200 -150 -100 0, 0000% -50 0 50 Simulated Magnitude 100 150 200 250

Error vs Simulated Magnitude (5% Meter Variance) 300, 0000% 250, 0000% Error 200, 0000% 150, 0000% TRUE FALSE 100, 0000% 50, 0000% 0 50 100 150 200 250 Simulated Magnitude 300 350 400

Overall Power vs. Simulated Magnitude (1% Process Variance) 1, 2 1 Overall Power 0, 8 0, 6 TRUE FALSE 0, 4 0, 2 0 0 20 40 60 80 Simulated Magnitude 100 120 140

Overall Power vs. Simulated Magnitude (3% Meter Variance) 1, 2 1 Overall Power 0, 8 0, 6 TRUE FALSE 0, 4 0, 2 -250 -200 -150 -100 0 -50 0 50 Simulated Magnitude 100 150 200 250

Overall Power vs. Simulated Magnitude (5% Meter Variance) 1, 2 1 Overall Power 0, 8 0, 6 TRUE FALSE 0, 4 0, 2 0 0 50 100 150 200 250 Simulated Magnitude 300 350 400

Simulation Procedure Energy balance without leak The pressure drop becomes: where In the presence of leak of magnitude b and location x from the head of the branch Miguel J. Bagajewicz and Emmanuel Cabrera. "Data Reconciliation in Gas Pipeline Systems. " Ind. Eng. Chem. Res 42, No. 22(2003): 1 -11

Simulation Procedure Problem formulation Without Error: With Error: Subject to: So:

Simulation Procedure Leak detection procedure: p Hypothesize leak in every branch and solve data reconciliation problem p Obtain GLR test statistic for each branch objno_leak –objwith_leak_k p Determine the maximum test statistic objno_leak - objwith_leak_k p We compare the max test statistic with the chosen threshold value: Max{objno_leak – objwith_leak_k}> threshold value: leak is identified and located in the branch corresponding to the maximum test statistic NOTE: Assuming only one possible error

Why pressure measurements? p Stream 1 and 2 measured p Discrepancy in flow measurement Sensor 1 Leak Sensor 2 Case 1 0. 4 0 0 Case 2 0 0. 4 0 Case 3 0 0 -0. 4

Sample Pipeline Network

Simulation Procedure - Leak in Pipe 1 Calculator Leak induced in Pipe 1 Optimizer

Simulation Results- Leak in Pipe 1 Leak Simulated Pipe 1 Location(m) 4000 Magnitude(kg/s) 4. 915 Measured Flow 15. 482 Measured Pressure (KPa) 2420. 3 Estimated Magnitude(kg/s) Estimated Location(m) 4. 640 4048 Pipe Best Objective function 1 15. 9834 2 18. 0199 3 60. 4256 4 60. 7056 5 21. 3695 6 16. 8630 7 78. 6864 8 81. 0650 9 123. 2020

Simulation Procedure - Leak in Pipe 8 Leak induced in Pipe 8

Simulation Results- Leak in Pipe 8 Leak Simulated Pipe 8 Location(m) 450 Magnitude(kg/s) 2. 611 Measured Flow 4. 946 Measured Pressure (k. Pa) 2160. 1 Estimated Magnitude(kg/s) Estimated Location(m) 2. 609 450 Pipe Best Objective function 1 126. 678 2 97. 438 3 101. 864 4 123. 710 5 126. 447 6 126. 447 7 63. 294 8 0. 151 9 159. 922

Simulation Procedure - Case Study p Case Study n n n p Perfect measurements Introduce noise in measurements Different leak magnitudes Error n n The error is calculated as follows: Error = (Calculated Value-True Value)/(True Value)

Simulation Procedure-Case Study Error vs. Leak Magnitude (Perfect Measurment) 1, 00% The system gives perfect results when perfect measurements are taken. Error 0, 50% 0, 00% 0 0, 5 1 1, 5 2 2, 5 -0, 50% -1, 00% Leak Magnitude 3 3, 5 4 4, 5

Simulation Procedure-Case Study Error VS. Leak Simulated As the simulated leak magnitude decreases, the error in the estimated magnitude increases. The Simulated magnitude has little effect on the location. There is no trend in identification of True/False values. 70, 00% 60, 00% 50, 00% Error 40, 00% Location True Magnitude True 30, 00% Location False Magnitude False 20, 00% 10, 00% 0 -10, 00% 0, 5 1 1, 5 2 Leak Simulated 2, 5 3 3, 5

Simulation Procedure-Case Study Overall Power VS. Leak Simulated 70 60 Overall Power 50 40 Overall Power True 30 As the simulated leak magnitude increases, the overall power increases. 20 10 0 0 0, 5 1 1, 5 2 Leak Simulated 2, 5 3 3, 5 Overall Power False

Future Research p. Run more simulations p. Explore multiple leak detection

Acknowledgements Miguel Bagajewicz p Quang Nguyen p Roman Voronov p Rufei Lu p

Questions ?